4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 by Delphix. All rights reserved.
29 #include <sys/fm/fs/zfs.h>
30 #include <sys/spa_impl.h>
31 #include <sys/nvpair.h>
33 #include <sys/fs/zfs.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/zfs_ioctl.h>
36 #include <sys/systeminfo.h>
37 #include <sys/sunddi.h>
38 #include <sys/zfeature.h>
45 * Pool configuration repository.
47 * Pool configuration is stored as a packed nvlist on the filesystem. By
48 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
49 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
50 * property set that allows them to be stored in an alternate location until
51 * the control of external software.
53 * For each cache file, we have a single nvlist which holds all the
54 * configuration information. When the module loads, we read this information
55 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
56 * maintained independently in spa.c. Whenever the namespace is modified, or
57 * the configuration of a pool is changed, we call spa_config_sync(), which
58 * walks through all the active pools and writes the configuration to disk.
61 static uint64_t spa_config_generation
= 1;
64 * This can be overridden in userland to preserve an alternate namespace for
65 * userland pools when doing testing.
67 char *spa_config_path
= ZPOOL_CACHE
;
68 int zfs_autoimport_disable
= 1;
71 * Called when the module is first loaded, this routine loads the configuration
72 * file into the SPA namespace. It does not actually open or load the pools; it
73 * only populates the namespace.
79 nvlist_t
*nvlist
, *child
;
86 if (zfs_autoimport_disable
)
91 * Open the configuration file.
93 pathname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
95 (void) snprintf(pathname
, MAXPATHLEN
, "%s%s",
96 (rootdir
!= NULL
) ? "./" : "", spa_config_path
);
98 file
= kobj_open_file(pathname
);
100 kmem_free(pathname
, MAXPATHLEN
);
102 if (file
== (struct _buf
*)-1)
105 if (kobj_get_filesize(file
, &fsize
) != 0)
108 buf
= kmem_alloc(fsize
, KM_SLEEP
);
111 * Read the nvlist from the file.
113 if (kobj_read_file(file
, buf
, fsize
, 0) < 0)
119 if (nvlist_unpack(buf
, fsize
, &nvlist
, KM_SLEEP
) != 0)
123 * Iterate over all elements in the nvlist, creating a new spa_t for
124 * each one with the specified configuration.
126 mutex_enter(&spa_namespace_lock
);
128 while ((nvpair
= nvlist_next_nvpair(nvlist
, nvpair
)) != NULL
) {
129 if (nvpair_type(nvpair
) != DATA_TYPE_NVLIST
)
132 VERIFY(nvpair_value_nvlist(nvpair
, &child
) == 0);
134 if (spa_lookup(nvpair_name(nvpair
)) != NULL
)
136 (void) spa_add(nvpair_name(nvpair
), child
, NULL
);
138 mutex_exit(&spa_namespace_lock
);
144 kmem_free(buf
, fsize
);
146 kobj_close_file(file
);
150 spa_config_write(spa_config_dirent_t
*dp
, nvlist_t
*nvl
)
155 int oflags
= FWRITE
| FTRUNC
| FCREAT
| FOFFMAX
;
160 * If the nvlist is empty (NULL), then remove the old cachefile.
163 err
= vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
168 * Pack the configuration into a buffer.
170 VERIFY(nvlist_size(nvl
, &buflen
, NV_ENCODE_XDR
) == 0);
172 buf
= vmem_alloc(buflen
, KM_SLEEP
);
173 temp
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
175 VERIFY(nvlist_pack(nvl
, &buf
, &buflen
, NV_ENCODE_XDR
,
178 #if defined(__linux__) && defined(_KERNEL)
180 * Write the configuration to disk. Due to the complexity involved
181 * in performing a rename from within the kernel the file is truncated
182 * and overwritten in place. In the event of an error the file is
183 * unlinked to make sure we always have a consistent view of the data.
185 err
= vn_open(dp
->scd_path
, UIO_SYSSPACE
, oflags
, 0644, &vp
, 0, 0);
187 err
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0,
188 UIO_SYSSPACE
, 0, RLIM64_INFINITY
, kcred
, NULL
);
190 err
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
192 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
195 (void) vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
199 * Write the configuration to disk. We need to do the traditional
200 * 'write to temporary file, sync, move over original' to make sure we
201 * always have a consistent view of the data.
203 (void) snprintf(temp
, MAXPATHLEN
, "%s.tmp", dp
->scd_path
);
205 err
= vn_open(temp
, UIO_SYSSPACE
, oflags
, 0644, &vp
, CRCREAT
, 0);
207 err
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0, UIO_SYSSPACE
,
208 0, RLIM64_INFINITY
, kcred
, NULL
);
210 err
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
212 err
= vn_rename(temp
, dp
->scd_path
, UIO_SYSSPACE
);
213 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
216 (void) vn_remove(temp
, UIO_SYSSPACE
, RMFILE
);
219 vmem_free(buf
, buflen
);
220 kmem_free(temp
, MAXPATHLEN
);
225 * Synchronize pool configuration to disk. This must be called with the
226 * namespace lock held. Synchronizing the pool cache is typically done after
227 * the configuration has been synced to the MOS. This exposes a window where
228 * the MOS config will have been updated but the cache file has not. If
229 * the system were to crash at that instant then the cached config may not
230 * contain the correct information to open the pool and an explicity import
234 spa_config_sync(spa_t
*target
, boolean_t removing
, boolean_t postsysevent
)
236 spa_config_dirent_t
*dp
, *tdp
;
239 boolean_t ccw_failure
;
242 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
244 if (rootdir
== NULL
|| !(spa_mode_global
& FWRITE
))
248 * Iterate over all cachefiles for the pool, past or present. When the
249 * cachefile is changed, the new one is pushed onto this list, allowing
250 * us to update previous cachefiles that no longer contain this pool.
252 ccw_failure
= B_FALSE
;
253 for (dp
= list_head(&target
->spa_config_list
); dp
!= NULL
;
254 dp
= list_next(&target
->spa_config_list
, dp
)) {
256 if (dp
->scd_path
== NULL
)
260 * Iterate over all pools, adding any matching pools to 'nvl'.
263 while ((spa
= spa_next(spa
)) != NULL
) {
265 * Skip over our own pool if we're about to remove
266 * ourselves from the spa namespace or any pool that
267 * is readonly. Since we cannot guarantee that a
268 * readonly pool would successfully import upon reboot,
269 * we don't allow them to be written to the cache file.
271 if ((spa
== target
&& removing
) ||
275 mutex_enter(&spa
->spa_props_lock
);
276 tdp
= list_head(&spa
->spa_config_list
);
277 if (spa
->spa_config
== NULL
||
278 tdp
->scd_path
== NULL
||
279 strcmp(tdp
->scd_path
, dp
->scd_path
) != 0) {
280 mutex_exit(&spa
->spa_props_lock
);
285 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
,
288 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
) {
289 VERIFY0(nvlist_lookup_string(spa
->spa_config
,
290 ZPOOL_CONFIG_POOL_NAME
, &pool_name
));
292 pool_name
= spa_name(spa
);
294 VERIFY(nvlist_add_nvlist(nvl
, pool_name
,
295 spa
->spa_config
) == 0);
296 mutex_exit(&spa
->spa_props_lock
);
299 error
= spa_config_write(dp
, nvl
);
301 ccw_failure
= B_TRUE
;
307 * Keep trying so that configuration data is
308 * written if/when any temporary filesystem
309 * resource issues are resolved.
311 if (target
->spa_ccw_fail_time
== 0) {
312 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE
,
313 target
, NULL
, NULL
, 0, 0);
315 target
->spa_ccw_fail_time
= gethrtime();
316 spa_async_request(target
, SPA_ASYNC_CONFIG_UPDATE
);
319 * Do not rate limit future attempts to update
322 target
->spa_ccw_fail_time
= 0;
326 * Remove any config entries older than the current one.
328 dp
= list_head(&target
->spa_config_list
);
329 while ((tdp
= list_next(&target
->spa_config_list
, dp
)) != NULL
) {
330 list_remove(&target
->spa_config_list
, tdp
);
331 if (tdp
->scd_path
!= NULL
)
332 spa_strfree(tdp
->scd_path
);
333 kmem_free(tdp
, sizeof (spa_config_dirent_t
));
336 spa_config_generation
++;
339 spa_event_notify(target
, NULL
, FM_EREPORT_ZFS_CONFIG_SYNC
);
343 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
344 * and we don't want to allow the local zone to see all the pools anyway.
345 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
346 * information for all pool visible within the zone.
349 spa_all_configs(uint64_t *generation
)
354 if (*generation
== spa_config_generation
)
357 VERIFY(nvlist_alloc(&pools
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
359 mutex_enter(&spa_namespace_lock
);
360 while ((spa
= spa_next(spa
)) != NULL
) {
361 if (INGLOBALZONE(curproc
) ||
362 zone_dataset_visible(spa_name(spa
), NULL
)) {
363 mutex_enter(&spa
->spa_props_lock
);
364 VERIFY(nvlist_add_nvlist(pools
, spa_name(spa
),
365 spa
->spa_config
) == 0);
366 mutex_exit(&spa
->spa_props_lock
);
369 *generation
= spa_config_generation
;
370 mutex_exit(&spa_namespace_lock
);
376 spa_config_set(spa_t
*spa
, nvlist_t
*config
)
378 mutex_enter(&spa
->spa_props_lock
);
379 nvlist_free(spa
->spa_config
);
380 spa
->spa_config
= config
;
381 mutex_exit(&spa
->spa_props_lock
);
385 * Generate the pool's configuration based on the current in-core state.
387 * We infer whether to generate a complete config or just one top-level config
388 * based on whether vd is the root vdev.
391 spa_config_generate(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int getstats
)
393 nvlist_t
*config
, *nvroot
;
394 vdev_t
*rvd
= spa
->spa_root_vdev
;
395 unsigned long hostid
= 0;
396 boolean_t locked
= B_FALSE
;
403 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
406 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
) ==
407 (SCL_CONFIG
| SCL_STATE
));
410 * If txg is -1, report the current value of spa->spa_config_txg.
413 txg
= spa
->spa_config_txg
;
416 * Originally, users had to handle spa namespace collisions by either
417 * exporting the already imported pool or by specifying a new name for
418 * the pool with a conflicting name. In the case of root pools from
419 * virtual guests, neither approach to collision resolution is
420 * reasonable. This is addressed by extending the new name syntax with
421 * an option to specify that the new name is temporary. When specified,
422 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
423 * to use the previous name, which we do below.
425 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
) {
426 VERIFY0(nvlist_lookup_string(spa
->spa_config
,
427 ZPOOL_CONFIG_POOL_NAME
, &pool_name
));
429 pool_name
= spa_name(spa
);
431 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
433 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
434 spa_version(spa
)) == 0);
435 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
437 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
438 spa_state(spa
)) == 0);
439 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
441 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
442 spa_guid(spa
)) == 0);
443 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
444 spa
->spa_errata
) == 0);
445 VERIFY(spa
->spa_comment
== NULL
|| nvlist_add_string(config
,
446 ZPOOL_CONFIG_COMMENT
, spa
->spa_comment
) == 0);
450 hostid
= zone_get_hostid(NULL
);
453 * We're emulating the system's hostid in userland, so we can't use
456 (void) ddi_strtoul(hw_serial
, NULL
, 10, &hostid
);
459 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_HOSTID
,
462 VERIFY0(nvlist_add_string(config
, ZPOOL_CONFIG_HOSTNAME
,
463 utsname()->nodename
));
466 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TOP_GUID
,
467 vd
->vdev_top
->vdev_guid
) == 0);
468 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_GUID
,
469 vd
->vdev_guid
) == 0);
470 if (vd
->vdev_isspare
)
471 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_IS_SPARE
,
474 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_IS_LOG
,
476 vd
= vd
->vdev_top
; /* label contains top config */
479 * Only add the (potentially large) split information
480 * in the mos config, and not in the vdev labels
482 if (spa
->spa_config_splitting
!= NULL
)
483 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
484 spa
->spa_config_splitting
) == 0);
488 * Add the top-level config. We even add this on pools which
489 * don't support holes in the namespace.
491 vdev_top_config_generate(spa
, config
);
494 * If we're splitting, record the original pool's guid.
496 if (spa
->spa_config_splitting
!= NULL
&&
497 nvlist_lookup_uint64(spa
->spa_config_splitting
,
498 ZPOOL_CONFIG_SPLIT_GUID
, &split_guid
) == 0) {
499 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_SPLIT_GUID
,
503 nvroot
= vdev_config_generate(spa
, vd
, getstats
, 0);
504 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
508 * Store what's necessary for reading the MOS in the label.
510 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
511 spa
->spa_label_features
) == 0);
513 if (getstats
&& spa_load_state(spa
) == SPA_LOAD_NONE
) {
514 ddt_histogram_t
*ddh
;
518 ddh
= kmem_zalloc(sizeof (ddt_histogram_t
), KM_SLEEP
);
519 ddt_get_dedup_histogram(spa
, ddh
);
520 VERIFY(nvlist_add_uint64_array(config
,
521 ZPOOL_CONFIG_DDT_HISTOGRAM
,
522 (uint64_t *)ddh
, sizeof (*ddh
) / sizeof (uint64_t)) == 0);
523 kmem_free(ddh
, sizeof (ddt_histogram_t
));
525 ddo
= kmem_zalloc(sizeof (ddt_object_t
), KM_SLEEP
);
526 ddt_get_dedup_object_stats(spa
, ddo
);
527 VERIFY(nvlist_add_uint64_array(config
,
528 ZPOOL_CONFIG_DDT_OBJ_STATS
,
529 (uint64_t *)ddo
, sizeof (*ddo
) / sizeof (uint64_t)) == 0);
530 kmem_free(ddo
, sizeof (ddt_object_t
));
532 dds
= kmem_zalloc(sizeof (ddt_stat_t
), KM_SLEEP
);
533 ddt_get_dedup_stats(spa
, dds
);
534 VERIFY(nvlist_add_uint64_array(config
,
535 ZPOOL_CONFIG_DDT_STATS
,
536 (uint64_t *)dds
, sizeof (*dds
) / sizeof (uint64_t)) == 0);
537 kmem_free(dds
, sizeof (ddt_stat_t
));
541 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
547 * Update all disk labels, generate a fresh config based on the current
548 * in-core state, and sync the global config cache (do not sync the config
549 * cache if this is a booting rootpool).
552 spa_config_update(spa_t
*spa
, int what
)
554 vdev_t
*rvd
= spa
->spa_root_vdev
;
558 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
560 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
561 txg
= spa_last_synced_txg(spa
) + 1;
562 if (what
== SPA_CONFIG_UPDATE_POOL
) {
563 vdev_config_dirty(rvd
);
566 * If we have top-level vdevs that were added but have
567 * not yet been prepared for allocation, do that now.
568 * (It's safe now because the config cache is up to date,
569 * so it will be able to translate the new DVAs.)
570 * See comments in spa_vdev_add() for full details.
572 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
573 vdev_t
*tvd
= rvd
->vdev_child
[c
];
574 if (tvd
->vdev_ms_array
== 0)
575 vdev_metaslab_set_size(tvd
);
576 vdev_expand(tvd
, txg
);
579 spa_config_exit(spa
, SCL_ALL
, FTAG
);
582 * Wait for the mosconfig to be regenerated and synced.
584 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
587 * Update the global config cache to reflect the new mosconfig.
589 if (!spa
->spa_is_root
)
590 spa_config_sync(spa
, B_FALSE
, what
!= SPA_CONFIG_UPDATE_POOL
);
592 if (what
== SPA_CONFIG_UPDATE_POOL
)
593 spa_config_update(spa
, SPA_CONFIG_UPDATE_VDEVS
);
596 #if defined(_KERNEL) && defined(HAVE_SPL)
597 EXPORT_SYMBOL(spa_config_sync
);
598 EXPORT_SYMBOL(spa_config_load
);
599 EXPORT_SYMBOL(spa_all_configs
);
600 EXPORT_SYMBOL(spa_config_set
);
601 EXPORT_SYMBOL(spa_config_generate
);
602 EXPORT_SYMBOL(spa_config_update
);
604 module_param(spa_config_path
, charp
, 0444);
605 MODULE_PARM_DESC(spa_config_path
, "SPA config file (/etc/zfs/zpool.cache)");
607 module_param(zfs_autoimport_disable
, int, 0644);
608 MODULE_PARM_DESC(zfs_autoimport_disable
, "Disable pool import at module load");